AI Drug Discovery That Learned the Rules: Tom Masterson, Deep Genomics

Tom Masterson, COO of Deep Genomics, on AI drug discovery where a hallucination can kill a patient. Deep Genomics gave its model a cholesterol target it had never seen, along with 3,500 candidate molecules. It picked nine, and one was a drug already approved and in patients. Tom explains why that means the model learned the rules instead of memorizing answers, how the company's lab-in-the-loop works, and why it designs intentionally toxic test molecules so the model learns what toxic looks like. Then the operating side: going from chief of staff to COO, getting scientists to adopt process, running waterfall and agile side by side, and the hardest question a biotech fiduciary faces. 🎧 Listen on your favorite podcast app: Apple Podcasts: https://podcasts.apple.com/us/podcast... Spotify: https://open.spotify.com/show/4OQY2oi... Newsletter: https://betweentwocoos.com ⏱ Chapters: 00:00 Cold open: no room for hallucination 00:27 An AI model picks an approved drug out of 3,500 01:13 Life sciences vs. healthcare, explained 02:25 Brendan Frey, Hinton's lab, and right but early 04:24 Transformers and the switch to foundation models 05:06 How a biological foundation model works 06:18 Retiring 40 task-specific models 08:02 GenomeKit and the engineering that survived 09:12 What lab in the loop means 10:09 Precision over creativity 11:58 Why pharma's historical data is biased 12:50 Pretend your code is going into a patient 13:19 Lab time vs code time 15:28 Raising money for a 10-year bet 17:30 Finding drugs the human brain couldn't 18:18 The siRNA result and why it matters 19:27 What siRNA is, in plain English 20:59 The model learned the rules 21:20 Brute force vs what nobody else can do 22:50 Tom's path: genetics, dental lasers, sales at 21 24:31 The push to get an MBA 25:11 Microbiome, a caregiver startup, and the pandemic 25:51 Joining Deep Genomics 26:55 Why genetic medicine mostly goes to the liver 28:01 Fixing ops at a university spinout 29:32 From chief of staff to COO 30:17 Getting scientists to adopt process 31:19 Eliminate yourself from the loop 32:38 A good time to be a generalist 33:39 Waterfall and agile side by side 35:39 The quarterly rhythm: portfolio, advisors, board 37:45 What the COO owns today 39:30 It is still a business 41:22 Personalized cancer vaccines and N of one 43:18 A hero to the world or to your shareholders 44:19 I never thought I'd see that: a two-week forensic audit 47:01 The barbecue champion 47:40 Wrap 📚 Mentioned in this episode: Deep Genomics: https://www.deepgenomics.com GenomeKit (open source): https://github.com/deepgenomics/Genom... Alnylam Pharmaceuticals: https://www.alnylam.com Moderna: https://www.modernatx.com Merck: https://www.merck.com Broad Institute: https://www.broadinstitute.org About Tom: Tom Masterson is chief operating officer of Deep Genomics, where he partners with the CEO on board and investor relations and turns the company's science into its business story. Before Deep Genomics, he was the first employee at Microbiome Insights and founded Caregiver Support. He holds a genetics degree from the University of British Columbia and an MBA from Harvard Business School. About Between Two COOs: Between Two COOs is the execution intelligence podcast for operating leaders. Host Michael Koenig sits down with COOs, presidents, and founders to talk about what it actually takes to run a company. From early-stage scaling to $50B operations. New episodes weekly. 🔗 Connect: Guest:   / mastersonthomas   Michael Koenig:   / michael-koenig514   Helm: https://helmapp.ai #BetweenTwoCOOs #COO #AIDrugDiscovery #Biotech